Overfitting in Trading

Overfitting (also called curve-fitting) is the process of creating a model so tailored to past data that it no longer captures underlying market structure — it has memorized the noise. In trading, this manifests as strategies that backtest brilliantly but lose money in forward testing.

**How overfitting happens:** 1. A trader starts with a simple strategy: "buy the 50-day moving average breakout" 2. Backtest shows a mediocre result (Sharpe 0.8) 3. Trader adjusts parameters: "what if it's the 47-day MA? The 52-day? What about adding a volume filter? RSI confirmation?" 4. After 200 parameter variations, one combination produces Sharpe 3.0 on historical data 5. The trader deploys this "optimized" strategy — and it immediately starts losing

**Symptoms of overfitting:** - Sharpe ratio above 3 in backtest (rare in legitimate strategies) - Strategy has more than 5–7 parameters - Win rate above 80% (markets rarely offer this in practice) - Maximum drawdown in backtest seems implausibly low - Small parameter changes cause large performance changes - Backtest equity curve is perfectly smooth

**Reducing overfitting:**

**1. Occam's Razor:** Prefer the simplest strategy that works. Two parameters > ten parameters if performance is similar.

**2. Out-of-sample testing:** Reserve data the strategy never sees during development. Test on it only once, after finalization.

**3. Walk-forward analysis:** Continuously test on unseen data across time.

**4. Monte Carlo simulation:** Randomize trade order and inputs to stress-test strategy robustness.

**5. Robustness testing:** Vary parameters by ±10–20%. If performance degrades sharply, the strategy is fragile and likely overfit.

**The "backtest fallacy" in crypto communities:** Many crypto content creators share "backtested strategies with 1000% returns." These are almost universally overfit. A fair test requires: realistic transaction costs, survivorship-bias-free asset selection, out-of-sample validation, and full cycle (bull + bear) data. Almost none of the viral crypto backtests meet these standards.